FyAgent: a local desktop control center for your AI coding tools

The problem FyAgent targets: configuration scattered across AI tools

If you run more than one AI coding tool, your settings live in more than one place. Each tool has its own idea of where a model provider is configured, where skills are installed, which MCP servers it can reach, and what system prompt it obeys. The README frames this as choices that are "originally scattered, hidden, and easy to change wrongly," and FyAgent's stated goal is to make them visible, editable, and portable.

The target user is someone already using AI agents, AI workers or assistants, not someone deciding whether to start. FyAgent does not give you a model. It gives you a place to point the tools you already have at one. The README deliberately avoids jargon: Provider, MCP and Prompt are described as the AI's brain source, tool connections and behavior instructions. That framing tells you who the documentation is written for. It is an operator's tool, not a framework for building agents.

How FyAgent works: scan, preview, write, verify

FyAgent is a Tauri desktop application with a Rust backend in src-tauri and a TypeScript frontend in src. That split matters for what it can do: the Rust side reads and writes configuration files on disk, and the frontend presents them.

The workflow described in the README has four beats. First, the AI software configuration area scans the machine for installed tools and lists QoderWork CN, TRAE Work CN, WorkBuddy, Grok Build, Codex, Claude Code and OpenCode. Second, you pick one and open its configuration page. Third, you edit models, Skills, MCP servers, prompts or authentication through whatever entry points that page offers. Fourth, you inspect a preview of the change scope before saving, then confirm or test the connection after saving.

That preview step is the design decision worth noting. Writing to another application's configuration is destructive if it goes wrong, and FyAgent treats the write as something to show you first. Whether the preview covers every field on every supported tool is not stated in the README; it says the preview exists and that you should check it.

Support is not uniform. The README's feature table says FyAgent scans all seven tools, but model and Provider editing applies to "the above software," Skills and MCP assignment only to "supported software," prompt management covers Grok Build, Codex, Claude Code, OpenCode, Gemini, OpenClaw and Hermes, and memory editing is limited to OpenClaw and Hermes long-term memory plus OpenClaw daily memory. Gemini, OpenClaw and Hermes appear in the prompt and memory rows without appearing in the scan row, so the per-tool matrix is the thing to read before you assume a feature applies to you.

Working data defaults to ~/.fyagent on the local machine. The README points to a configuration file document for exact write locations and backup methods rather than listing them inline.

Installing FyAgent and configuring one AI tool

FyAgent ships as a desktop installer, not a package you add to a project. The README directs you to GitHub Releases and names the artifacts: FyAgent-X.Y.Z-macOS.dmg on macOS, and FyAgent-X.Y.Z-Windows-x64-setup.exe or FyAgent-X.Y.Z-Windows-arm64-setup.exe on Windows. Windows currently offers an NSIS installer only; the README states there is no MSI and no portable ZIP.

The quick start in the README is four steps:

# 1. Download the installer for your system from GitHub Releases
# 2. Open "AI software configuration" and scan for installed tools
# 3. Pick a tool and open its configuration page
# 4. Check the preview before saving, then confirm or test the connection

Those are the README's steps in its own order, not commands to paste. The only literal commands in the README are for local development, which requires mise >= 2026.8.6 installed globally:

mise trust
mise run bootstrap
mise run system:check
mise run dev

mise run build produces a build for the current system, and mise run check is the pre-commit gate. The development documentation covers the full toolchain and layered checks.

Before installing, the README asks you to read the release notes for that version and verify the checksum, signing-status.json and build provenance. It makes a point worth repeating: a NotSigned status only describes signature state and does not by itself prove a file is safe. macOS builds use Apple Developer ID signing and Apple notarization. Windows builds are not described as signed in the README.

A first real use looks like this: scan, open Codex or Claude Code, change the model or Provider, look at the preview, save, then test the connection from the page. If the tool you care about is not in the scan list, FyAgent has nothing to configure for it.

Where FyAgent is the wrong tool

The licence is the first boundary. FyAgent is source-available, not open source as OSI defines it. Its own components and modifications use PolyForm Noncommercial License 1.0.0, and commercial use requires separate written authorization. Parts inherited from CC Switch remain under MIT. If you are evaluating this for a company, that is a procurement question before it is a technical one, and the repository carries LICENSING.md, COMMERCIAL-LICENSE.md and THIRD_PARTY_NOTICES.md for that purpose. Note that the repository's declared license field is NOASSERTION, which is consistent with a split-licence codebase rather than a single SPDX identifier.

Second, FyAgent manages other tools; it does not replace them. If you use one AI tool and are happy editing its config by hand, the scan-preview-write loop adds a layer without removing work.

Third, the README carries its own warning: FyAgent is still under development, and you should back up important configuration before upgrading. That is an explicit statement that upgrades can affect your settings. The README does not document rollback, and it does not document a downgrade path between releases.

Fourth, platform reach is Windows and macOS. Linux is not listed as a supported platform in the README's badge or its installer names, even though the project topics include cross-platform and the frontend stack would run elsewhere.

Finally, the supported-software list is a closed set of seven tools for scanning. Anything outside it, including tools that store configuration in the same formats, is not handled.

FyAgent versus editing each tool's config yourself, or CC Switch

The honest alternative is doing nothing: edit each tool's own configuration files directly. The difference is not capability, because FyAgent is writing those same files. The difference is that FyAgent shows you a preview of the change scope before it writes, and gives you one screen per tool instead of one file format per tool. If you already know where every file lives and you change settings rarely, direct editing is fewer moving parts and has no licence question.

The closer comparison is CC Switch, because FyAgent's current desktop application is described as evolving from it, and inherited code keeps its original copyright and MIT licence. The README does not enumerate what FyAgent adds on top, so the practical difference to verify is scope: FyAgent's feature table covers scanning seven AI tools, model and Provider editing, Skills, MCP, prompts and memory files, whereas the README only tells you that CC Switch is the base. Anyone choosing between them should read the release notes and the licensing files rather than assume the newer name is a strict superset.

A third option is per-tool profile managers built into individual AI tools. Those stay inside one tool's boundary by construction, so they cannot give you one view across Codex, Claude Code and OpenCode. FyAgent's whole premise is that the cross-tool view is the product.

Maintenance cadence, upgrade cost and licence implications

The repository is not archived, and the last push was on 2026-09-10. Releases are frequent: v0.4.4 on 2026-09-07, v0.4.3 on 2026-09-01, v0.4.2 on 2026-08-20. Three releases in under three weeks at the 0.4 line means the surface is still moving, and the README's own release-status note says the project is still under development and tells you to back up important configuration before upgrading.

That combination sets the upgrade cost. Every release may change how configuration is written to other tools. The README does not document rollback, so backing up ~/.fyagent and the affected tools' configuration files before an upgrade is the only stated mitigation. If you manage several machines, that backup step multiplies.

The build pipeline is heavier than a typical side project: mise for toolchain tasks, pnpm 10.12.3 as the package manager, a Rust toolchain pinned in rust-toolchain.toml, Python >= 3.14,<3.15 via uv for development, and a dedicated dmg-layout dependency group for macOS packaging. Building from source means installing and trusting all of that.

On licensing, the practical reading is that personal and other noncommercial use is what PolyForm Noncommercial 1.0.0 permits, and any commercial use requires written authorization obtained separately. This is a description of what the repository states, not legal advice; read LICENSING.md, LICENSE, LICENSES/ and THIRD_PARTY_NOTICES.md and get your own counsel if you plan to use it at work.

Editorial conclusion

FyAgent fits engineers who already run several AI coding tools on one machine and want their model, Skills, MCP, prompt and memory settings visible and editable instead of scattered across hidden config files. It does not fit commercial use without a written licence, or anyone expecting an OSI open source project: the project's own components are PolyForm Noncommercial 1.0.0. Before installing, verify the release against the checksum, signing-status.json and build provenance the README asks you to check, and confirm the specific tool you use appears in the supported list, since support differs per software.

Frequently asked questions

Official sources

1. fy-agent/fyagent on GitHub
2. Issues
3. Project website
4. README
5. Releases